Add zstd relevant code to a separate PR

Author: Gabriel Tincu

kafka/codec.py

@@ -16,6 +16,11 @@
 except ImportError:
     snappy = None
 
+try:
+    import zstandard as zstd
+except ImportError:
+    zstd = None
+
 try:
     import lz4.frame as lz4
 
@@ -58,6 +63,10 @@ def has_snappy():
     return snappy is not None
 
 
+def has_zstd():
+    return zstd is not None
+
+
 def has_lz4():
     if lz4 is not None:
         return True
@@ -299,3 +308,15 @@ def lz4_decode_old_kafka(payload):
         payload[header_size:]
     ])
     return lz4_decode(munged_payload)
+
+
+def zstd_encode(payload):
+    if not zstd:
+        raise NotImplementedError("Zstd codec is not available")
+    return zstd.ZstdCompressor().compress(payload)
+
+
+def zstd_decode(payload):
+    if not zstd:
+        raise NotImplementedError("Zstd codec is not available")
+    return zstd.ZstdDecompressor().decompress(payload)

kafka/producer/kafka.py

@@ -12,7 +12,7 @@
 
 import kafka.errors as Errors
 from kafka.client_async import KafkaClient, selectors
-from kafka.codec import has_gzip, has_snappy, has_lz4
+from kafka.codec import has_gzip, has_snappy, has_lz4, has_zstd
 from kafka.metrics import MetricConfig, Metrics
 from kafka.partitioner.default import DefaultPartitioner
 from kafka.producer.future import FutureRecordMetadata, FutureProduceResult
@@ -119,7 +119,7 @@ class KafkaProducer(object):
                 available guarantee.
             If unset, defaults to acks=1.
         compression_type (str): The compression type for all data generated by
-            the producer. Valid values are 'gzip', 'snappy', 'lz4', or None.
+            the producer. Valid values are 'gzip', 'snappy', 'lz4', 'zstd' or None.
             Compression is of full batches of data, so the efficacy of batching
             will also impact the compression ratio (more batching means better
             compression). Default: None.
@@ -339,6 +339,7 @@ class KafkaProducer(object):
         'gzip': (has_gzip, LegacyRecordBatchBuilder.CODEC_GZIP),
         'snappy': (has_snappy, LegacyRecordBatchBuilder.CODEC_SNAPPY),
         'lz4': (has_lz4, LegacyRecordBatchBuilder.CODEC_LZ4),
+        'zstd': (has_zstd, DefaultRecordBatchBuilder.CODEC_ZSTD),
         None: (lambda: True, LegacyRecordBatchBuilder.CODEC_NONE),
     }
 
@@ -388,6 +389,9 @@ def __init__(self, **configs):
         if self.config['compression_type'] == 'lz4':
             assert self.config['api_version'] >= (0, 8, 2), 'LZ4 Requires >= Kafka 0.8.2 Brokers'
 
+        if self.config['compression_type'] == 'zstd':
+            assert self.config['api_version'] >= (2, 1, 0), 'Zstd Requires >= Kafka 2.1.0 Brokers'
+
         # Check compression_type for library support
         ct = self.config['compression_type']
         if ct not in self._COMPRESSORS:

kafka/record/default_records.py

@@ -62,8 +62,8 @@
 )
 from kafka.errors import CorruptRecordException, UnsupportedCodecError
 from kafka.codec import (
-    gzip_encode, snappy_encode, lz4_encode,
-    gzip_decode, snappy_decode, lz4_decode
+    gzip_encode, snappy_encode, lz4_encode, zstd_encode,
+    gzip_decode, snappy_decode, lz4_decode, zstd_decode
 )
 import kafka.codec as codecs
 
@@ -97,6 +97,7 @@ class DefaultRecordBase(object):
     CODEC_GZIP = 0x01
     CODEC_SNAPPY = 0x02
     CODEC_LZ4 = 0x03
+    CODEC_ZSTD = 0x04
     TIMESTAMP_TYPE_MASK = 0x08
     TRANSACTIONAL_MASK = 0x10
     CONTROL_MASK = 0x20
@@ -111,6 +112,8 @@ def _assert_has_codec(self, compression_type):
             checker, name = codecs.has_snappy, "snappy"
         elif compression_type == self.CODEC_LZ4:
             checker, name = codecs.has_lz4, "lz4"
+        elif compression_type == self.CODEC_ZSTD:
+            checker, name = codecs.has_zstd, "zstd"
         if not checker():
             raise UnsupportedCodecError(
                 "Libraries for {} compression codec not found".format(name))
@@ -185,6 +188,8 @@ def _maybe_uncompress(self):
                     uncompressed = snappy_decode(data.tobytes())
                 if compression_type == self.CODEC_LZ4:
                     uncompressed = lz4_decode(data.tobytes())
+                if compression_type == self.CODEC_ZSTD:
+                    uncompressed = zstd_decode(data)
                 self._buffer = bytearray(uncompressed)
                 self._pos = 0
         self._decompressed = True
@@ -517,6 +522,8 @@ def _maybe_compress(self):
                 compressed = snappy_encode(data)
             elif self._compression_type == self.CODEC_LZ4:
                 compressed = lz4_encode(data)
+            elif self._compression_type == self.CODEC_ZSTD:
+                compressed = zstd_encode(data)
             compressed_size = len(compressed)
             if len(data) <= compressed_size:
                 # We did not get any benefit from compression, lets send

kafka/record/memory_records.py

@@ -117,7 +117,7 @@ class MemoryRecordsBuilder(object):
 
     def __init__(self, magic, compression_type, batch_size):
         assert magic in [0, 1, 2], "Not supported magic"
-        assert compression_type in [0, 1, 2, 3], "Not valid compression type"
+        assert compression_type in [0, 1, 2, 3, 4], "Not valid compression type"
         if magic >= 2:
             self._builder = DefaultRecordBatchBuilder(
                 magic=magic, compression_type=compression_type,

test/test_codec.py

@@ -7,11 +7,12 @@
 from kafka.vendor.six.moves import range
 
 from kafka.codec import (
-    has_snappy, has_lz4,
+    has_snappy, has_lz4, has_zstd,
     gzip_encode, gzip_decode,
     snappy_encode, snappy_decode,
     lz4_encode, lz4_decode,
     lz4_encode_old_kafka, lz4_decode_old_kafka,
+    zstd_encode, zstd_decode,
 )
 
 from test.testutil import random_string
@@ -113,3 +114,11 @@ def test_lz4_incremental():
         b2 = lz4_decode(lz4_encode(b1))
         assert len(b1) == len(b2)
         assert b1 == b2
+
+
+@pytest.mark.skipif(not has_zstd(), reason="Zstd not available")
+def test_zstd():
+    for _ in range(1000):
+        b1 = random_string(100).encode('utf-8')
+        b2 = zstd_decode(zstd_encode(b1))
+        assert b1 == b2

test/test_producer.py

@@ -23,7 +23,7 @@ def test_buffer_pool():
 
 
 @pytest.mark.skipif(not env_kafka_version(), reason="No KAFKA_VERSION set")
-@pytest.mark.parametrize("compression", [None, 'gzip', 'snappy', 'lz4'])
+@pytest.mark.parametrize("compression", [None, 'gzip', 'snappy', 'lz4', 'zstd'])
 def test_end_to_end(kafka_broker, compression):
 
     if compression == 'lz4':
@@ -34,10 +34,15 @@ def test_end_to_end(kafka_broker, compression):
         elif platform.python_implementation() == 'PyPy':
             return
 
+    if compression == 'zstd' and env_kafka_version() < (2, 1, 0):
+        return
+    env_version = env_kafka_version()
+    api_version = env_version if env_version >= (2, 1, 0) else None
     connect_str = ':'.join([kafka_broker.host, str(kafka_broker.port)])
     producer = KafkaProducer(bootstrap_servers=connect_str,
                              retries=5,
                              max_block_ms=30000,
+                             api_version=api_version,
                              compression_type=compression,
                              value_serializer=str.encode)
     consumer = KafkaConsumer(bootstrap_servers=connect_str,
@@ -81,16 +86,22 @@ def test_kafka_producer_gc_cleanup():
 
 
 @pytest.mark.skipif(not env_kafka_version(), reason="No KAFKA_VERSION set")
-@pytest.mark.parametrize("compression", [None, 'gzip', 'snappy', 'lz4'])
+@pytest.mark.parametrize("compression", [None, 'gzip', 'snappy', 'lz4', 'zstd'])
 def test_kafka_producer_proper_record_metadata(kafka_broker, compression):
+    if compression == 'zstd' and env_kafka_version() < (2, 1, 0):
+        return
     connect_str = ':'.join([kafka_broker.host, str(kafka_broker.port)])
+    env_version = env_kafka_version()
+    api_version = env_version if env_version >= (2, 1, 0) else None
     producer = KafkaProducer(bootstrap_servers=connect_str,
                              retries=5,
+                             api_version=api_version,
                              max_block_ms=30000,
                              compression_type=compression)
     magic = producer._max_usable_produce_magic()
 
     # record headers are supported in 0.11.0
+
     if env_kafka_version() < (0, 11, 0):
         headers = None
     else:

tox.ini

@@ -15,6 +15,7 @@ deps =
     pytest-mock
     mock
     python-snappy
+    zstandard
     lz4
     xxhash
     crc32c